Supervised Grid-of-Tries: A Novel Framework for Classifier Management

Abstract. Packet classification is the problem of identifying which one of a set of rules maintained in a database is best matched by an incoming packet at a router and taking the action specified by the rule. This is a uniform enabler for many new network services like firewalls, quality of service and virtual private networks. These services require dynamic management of rules. While many algorithms recently proposed can perform packet classification at very high speeds, rule update times for these are not very fast. This paper presents an efficient classifier management algorithm, which effectively reduces the rule update time for the well known Grid-of-Tries classifier. To this end, we have devised a novel structure called Supervised Grid-of-Tries, which employs additional tracking pointers embedded into the trie to facilitate efficient rule updates